Hala Abualsaud – habualsa@ucsd.edu
LinkedIn: linkedin.com/in/hala-abualsaud
ECE, University of California San Diego
San Diego, California 92130
United States
Peter Gerstoft – pgerstoft@ucsd.edu.
ECE, University of California San Diego
San Diego, California 92130
United States
Popular version of 2aCA7 – Acoustic Simultaneous Localization and Mapping for Drone Navigation in Complex Environments
Presented at the 188th ASA Meeting
Read the abstract at https://eppro01.ativ.me/appinfo.php?page=Session&project=ASAICA25&id=3867163&server=eppro01.ativ.me
–The research described in this Acoustics Lay Language Paper may not have yet been peer reviewed–
When drones fly indoors, inside warehouses, tunnels, or disaster zones, they can’t rely on GPS or cameras to know where they are. Instead, we propose something different: a drone that “listens” to its surroundings to navigate and map the environment.
We developed a new system called acSLAM (acoustic Simultaneous Localization and Mapping) that uses sound to guide a drone in 3D space. Our drone carries three microphone arrays, arranged in triangles, along with a motion sensor called an IMU (inertial measurement unit). As the drone moves, it records sounds and small changes in movement. Using this information, it estimates its own position and finds where multiple sound sources are located at the same time.
To handle the complexity of real 3D motion (where rotations can easily become unstable), we represent the drone’s orientation using quaternions – a way of describing rotation that avoids problems like gimbal lock, where the drone would otherwise lose its sense of direction. Quaternions work better than traditional methods because they keep track of rotation smoothly and consistently, even during fast or complex motion. They don’t get tripped up by tricky angles or repeated turns, which helps the drone stay accurately oriented as it moves through 3D space.
Our system works by first listening for where sounds are coming from (their angle of arrival) and measuring time differences (time difference of arrival) between microphones. Combining these clues with the drone’s movement, acSLAM builds a map of where sounds are in the room, like where people are talking or where machines are running.
We use advanced filtering methods (particle filters for the drone’s movement and Extended Kalman filters for the sound sources) to make sense of noisy real-world data. The system updates itself every step of the way, refining the drone’s position and improving the map as it gathers more information.
In testing, we found that using multiple sound observations, instead of relying on just one dramatically improved the drone’s ability to localize itself and map sources accurately. Even when the drone made sharp turns or accelerated quickly, the system stayed reliable.
This approach has exciting applications: drones could someday explore collapsed buildings, find survivors after disasters, or inspect underground spaces — all by listening carefully to their environment, without needing light, cameras, or external signals.
In short, we taught a drone not just to hear but to think about what it hears, and fly smarter because of it.